Environmental Science & Technology - ACS Publications - American

Conventional prescriptive emissions controls or cap-and-trade programs consider all emissions similarly regardless of when they occur, despite the fac...
0 downloads 0 Views 1MB Size
Article pubs.acs.org/est

Flexible NOx Abatement from Power Plants in the Eastern United States Lin Sun,†,§ Mort Webster,† Gary McGaughey,‡ Elena C. McDonald-Buller,*,‡ Tammy Thompson,†,‡ Ronald Prinn,§ A. Denny Ellerman,⊥ and David T. Allen‡ †

Engineering Systems Division, §Department of Earth, Atmosphere and Planetary Sciences, and ⊥Sloan School of Management, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United States ‡ Center for Energy and Environmental Resources, The University of Texas at Austin, Austin, Texas 78758, United States S Supporting Information *

ABSTRACT: Emission controls that provide incentives for maximizing reductions in emissions of ozone precursors on days when ozone concentrations are highest have the potential to be cost-effective ozone management strategies. Conventional prescriptive emissions controls or cap-and-trade programs consider all emissions similarly regardless of when they occur, despite the fact that contributions to ozone formation may vary. In contrast, a time-differentiated approach targets emissions reductions on forecasted high ozone days without imposition of additional costs on lower ozone days. This work examines simulations of such dynamic air quality management strategies for NOx emissions from electric generating units. Results from a model of day-specific NOx pricing applied to the Pennsylvania−New Jersey−Maryland (PJM) portion of the northeastern U.S. electrical grid demonstrate (i) that sufficient flexibility in electricity generation is available to allow power production to be switched from high to low NOx emitting facilities, (ii) that the emission price required to induce EGUs to change their strategies for power generation are competitive with other control costs, (iii) that dispatching strategies, which can change the spatial and temporal distribution of emissions, lead to ozone concentration reductions comparable to other control technologies, and (iv) that air quality forecasting is sufficiently accurate to allow EGUs to adapt their power generation strategies.



INTRODUCTION Elevated concentrations of ground-level ozone are among the most persistent and pervasive air quality concerns in the United States (U.S.). Ground-level ozone is a secondary pollutant formed by photochemical reactions of oxides of nitrogen (NOx) and volatile organic compounds (VOCs). National Ambient Air Quality Standards (NAAQS) for ground-level ozone have become increasingly stringent over the past several decades, with significant changes in averaging time, level, and form. Although significant improvements in air quality have occurred across the eastern U.S. over the past two decades,1−3 elevated ozone concentrations remain a regional concern. The Northeast States for Coordinated Air Use Management4 found that the most severe episodes are associated with the passage of slow-moving, high-pressure systems during the summer that extend from the Midwest to the middle or southern Atlantic states. Clear skies and high temperatures that favor ozone formation and a circulation pattern that favors long-range pollution transport can lead to ozone episodes of strong intensity and long duration. Reductions of NOx emissions, in © 2012 American Chemical Society

particular from electric generating units (EGUs), have been critical for achieving regional ozone reductions, while reductions of VOC emissions alone or in combination with NOx emissions reductions have been important in local circumstances.4 NOx emissions from EGUs represented 20% (3.7 million tons) of total U.S. anthropogenic NOx emissions in 2005, behind on-road mobile sources (36%) and nonroad equipment (23%).5 Prescriptive reductions of NOx emissions from EGUs can be achieved by combustion modifications, such as low NOx burners and boiler optimization, and postcombustion technologies, including selective noncatalytic reduction (SNCR) and selective catalytic reduction (SCR). SCR removes nitrogen oxides through reactions with ammonia in a catalyst bed. SNCR also uses ammonia or another reducing agent but without the Received: Revised: Accepted: Published: 5607

December March 20, March 20, March 20,

7, 2011 2012 2012 2012

dx.doi.org/10.1021/es204290s | Environ. Sci. Technol. 2012, 46, 5607−5615

Environmental Science & Technology

Article

This work focuses on the effects of time-differentiated NOx regulation of EGUs on ozone concentrations in the eastern U.S. Time-differentiated trading lowers NOx emissions from EGUs even further on days with forecasted elevated ozone by utilizing pricing incentives for electricity redispatching. The incremental cost of this strategy is the added cost of operating more expensive units (less efficient or higher fuel cost), but this cost is only incurred on the targeted high ozone days. In contrast, a prescriptive approach mandating additional controls adds both capital and operating costs for each unit receiving controls. A major technical barrier to a time-differentiated approach is the necessary reliance on weather and atmospheric chemistry forecasting, which is generally perceived to have large uncertainty.30−33 False positive events (a forecasted high ozone day that in actuality is a low ozone day) may increase costs unnecessarily, while false negative events (a forecasted low ozone day that in actuality is a high ozone day) decrease the amount of ozone that could otherwise be reduced. In this work, the air quality benefits and cost-effectiveness of timedifferentiated trading are compared with SCR and SNCR, postcombustion technologies likely to be implemented in the future. Although these simulations considered time-varying NOx pricing, such a system could be implemented as a timevarying redemption ratio for allowances under a cap, i.e., on designated high ozone days instead of surrendering a single allowance for each ton of NOx emitted, the regulatory authority would require that a number of allowances greater than one be surrendered for each ton.

catalyst. SCR is more effective, with up to 90% removal, but at a higher cost than SNCR, which achieves 30% to 80% removal depending on the specific unit. Recognition of the contribution of NOx emissions from EGUs to the regional formation and transport of ozone and fine particulate matter in the eastern U.S. prompted programs such as the NOx SIP Call in 1998 that required reductions of emissions from power plants and large combustion sources in 22 states and the District of Columbia.4,6−12 The Clean Air Interstate Rule (CAIR) issued in 2005 included a NOx control program aimed at a 28-state region in the eastern U.S.13 The U.S. EPA14,15 finalized the Cross-State Air Pollution Rule (CSAPR) as a likely replacement to CAIR for achieving reductions of NOx and sulfur dioxide (SO2) from power plants in the eastern half of the U.S in July 2011. A map of affected states under the CSAPR is given in Section S1, Supporting Information.14 For affected states based on 2007 emissions data contained in the U.S. EPA’s 2010 Emissions & Generation Resource Integrated Database (eGRID2010), approximately 84% of coal-fired utility boilers are already equipped with combustion controls, whereas SNCR and SCR are installed on only 9% and 20% of boilers. Emissions cap and trading has become the preferred federal policy instrument for achieving reductions of NOx and SO2 from EGUs.16 The NOx Budget Trading Program of the NOx SIP Call,16,17 its predecessor the Ozone Transport Commission’s NOx Budget Program,18 and the U.S. Acid Rain Program2 have been among the most prominent of these programs. The recently issued CSAPR will allow both intraand interstate emissions trading with provisions ensuring that states achieve pollution reductions.14,15 Generally, in emissions cap and trading programs, facilities are provided or purchase emission allowances; facilities that reduce their emissions below their allowances can trade or sell allowances in a market, and total emissions are reduced by lowering the number of allowances (the cap) over time. Routine emissions data are obtained from Continuous Emissions Monitoring Systems (CEMS) installed at affected facilities. An extensive body of previous studies has documented the successes as well as the challenges and concerns for these programs.16,19−26 Market-based NOx cap and trade programs have historically treated emissions as equivalent regardless of where or when they occur. Mauzerall et al.27 found that the amount of ozone formed by the same quantity of NOx could vary by a factor of 5 under different meteorological conditions in the eastern U.S, and that variation in health damages depended strongly on the size of the exposed population. Mauzerall et al.27 and Tong et al.28 suggested that incentives for reducing NOx emissions at times and locations where health damages are greatest could provide more effective reduction of damages from emissions permitted under a cap. Recent analyses of the use of NOx emissions allocations by EGUs by Martin29 in the northeastern U.S. have suggested that facilities disproportionately reduce NOx emissions early in the summer, conserving allocations for later in the season when peak demand in July and August result in higher electricity prices, even though this is also the period during which elevated ozone concentrations are more frequent. In addition, states preparing attainment plans will be challenged with the higher marginal costs of additional permanent or seasonal emissions reductions in the future if the NAAQS for ozone are revised. These findings suggest the utility of considering differentiated NOx regulation.



METHODS Simulation of the Electrical Power Grid. In the eastern U.S., most power plants that participate in the seasonal NOx cap-and-trade programs also participate in one of three wholesale electricity markets: the New England Power Pool, the New York Power Pool, or the Pennsylvania−New Jersey− Maryland (PJM) Interconnect. The Classic PJM Interconnect, selected for this work, included Pennsylvania−New Jersey− Maryland as well as the District of Columbia and Delaware. PJM has since expanded to include 10 additional states and is currently the world’s largest competitive wholesale electricity market. It is a federally regulated regional transmission organization (RTO) that coordinates buying and selling through central dispatching. An optimal power flow (OPF) model was used to simulate generation dispatching accounting for transmission constraints, incremental generation costs, and NOx emissions under different NOx pricing scenarios in the Classic PJM grid based on previous work by Martin.29 The model represented the set of EGUs as given in EPA’s 2005 Emissions & Generation Resource Integrated Database (eGRID) database and the detailed transmission system obtained from the Independent System Operators (ISOs). Coal-fired facilities accounted for 48% of the electricity generation in Classic PJM, followed by nuclear (33%) and gas (8%) facilities. Section S2, Supporting Information, presents annual NOx emissions for 280 Classic PJM facilities from the 2005 eGRID database. Total NOx emissions were 2.7 × 105 tons/yr with 91% attributed to coalfired facilities. The PJM Financial Transmission Rights (FTR) base-case power flows simulated hours with average electricity demand, while the PJM FTR model was scaled to approximate the higher demand hours. Generation costs were estimated using plant-level data in eGRID and calibrated to reproduce historical 5608

dx.doi.org/10.1021/es204290s | Environ. Sci. Technol. 2012, 46, 5607−5615

Environmental Science & Technology

Article

could be implemented using an allowance trading system, but on high ozone days, increasing the redemption ratio (allowances surrendered per ton emitted) to a value greater than unity. The measure of abatement costs, as described above, is the increased cost of generation from substituting sources with lower emissions. Incremental costs of NOx emissions reductions below base case ($2k/ton) emissions were calculated for the SNCR and SCR scenarios as the levelized costs for the capital expenditure plus variable operating costs. Cost calculations for both technology scenarios are presented in detail in Section S3, Supporting Information. An average daily cost of $0.98 ± 0.12 M/day during the June−Sept ozone season in 2005 dollars was determined for SNCR assuming that all coal-fired facilities install the technology. SCRs were assumed to be installed at 32% of coal-fired EGUs that had the highest NOx emission rates in Classic PJM. The proportion of coal-fired EGUs that receive SCRs were calibrated such that the $100k/ton trading case scenario and the SCR scenario had equivalent average daily abatement costs under an ozone concentration threshold of 75 ppb. This design allowed the reductions from similar cost scenarios to be compared. Controls were added to the highest NOx emission rate units first until the costs were comparable. Because the time-differentiated scenarios assumed no costs outside of the ozone season, SCR costs were assumed to be incurred only during the ozone season for parallel comparison. The average daily SCR cost for the subset of coal plants was $1.06 ± 0.35M/day. Simulation of Ozone Impacts. The Comprehensive Air Quality Model with Extensions (CAMx) version 4.51 (http:// www.camx.com) was used to examine the effects of NOx emissions changes on ground-level ozone concentrations. The model was developed by ENVIRON36 based on data from the Central Regional Air Planning Association (CENRAP) for annual modeling of regional haze and visibility in the eastern U.S. (http://www.epa.gov/scram001/reports/nox_sip.pdf). Meteorological data were developed by the Iowa Department of Natural Resources (IDNR) using the Pennsylvania State University (PSU)/National Center for Atmospheric Research (NCAR) Mesoscale Model Version 5 (MM5).37 Two broad types of analyses were conducted: deterministic and stochastic. All scenarios differed only in their NOx emissions from EGUs. Three deterministic simulations were conducted to examine the effects of increases in NOx pricing ($30k, $50k, or $100k/ton) implemented in Classic PJM on all days of the ozone season. Three stochastic scenarios were conducted to examine the effects of increases in NOx pricing ($30k/ton, $50k/ton, or $100k/ton) implemented only on high ozone days in Classic PJM. Results for the deterministic and stochastic scenarios were compared to the base case simulation with a NOx price of $2k/ ton and to simulations with SNCR or SCR installed on all or a subset of coal plants, on all days of the ozone season. The horizontal grid configuration is shown in Section S4, Supporting Information, along with the locations of 37 Philadelphia/Baltimore area ozone ground monitoring stations. Simulated daily maximum 8-h averaged ozone concentrations in the Philadelphia/Baltimore region were used in the stochastic analysis described below. Stochastic Analysis. A stochastic model integrated the results from the power system and photochemical models. Rather than using observed ozone concentrations, the modeled meteorology and ozone impacts for all days in the summer of 2002 were used as the population from which to sample future

observations. The model determined hourly generation in each unit to minimize cost, subject to transmission constraints and security contingencies. The OPF software used was PowerWorld Version 13. The optimal power flow model is a nonlinear optimization algorithm that minimizes total (fixed plus variable) operating cost subject to a set of equality and inequality constraints, including meeting demand, enforcing transmission line constraints, generator unit minimum and maximum power levels, and accounting for line losses. Instead of gradient-based and Newton solution methods, a linear programming (LP) approach was used, which allowed the inclusion of inequality constraints.34 The basic approach was iterative, solving the load flow problem,35 creating a linear objective function and linearizing constraints for those results, then solving the primal LP to obtain an improved solution for power output at each unit. Cost Calculations. The cost of a time-differentiated policy was calculated as the change in variable cost of electricity generation under a different emissions price relative to the base case. The change in the electricity generation cost for an individual EGU was a function of the generator’s output level, the unit’s heat rate, the cost of fuel, the unit’s emission rate, and the allowance price. In the OPF model, variable costs of the power plants were represented by linear cost curves, in which increased emissions were incorporated as an additional fuel cost: ci($/MWh) = Hi(pfi + pni Ni) + O&Mi

(1)

where for each generating unit i, Hi is its heat rate (mmBTU/ MWh), pfi is the price of fuel ($/mmBTU), pni is the price of the pollutant permits ($/ton), Ni is the unit’s emission rate (short tons/mmBTU), and O&Mi is the unit’s variable operation and maintenance costs ($/MWh). Data on the average delivered cost of fuel for natural gas, coal, petroleum products, and petroleum coke delivered to the electricity sector from the EIA’s Electric Power Monthly for August 2005 were used to generate the cost curves. These data were matched to the generating units by state and fuel. Variable O&M data were from the Annual Energy Outlook for 2006 matched roughly by technology type and fuel. The U.S. EPA CEMS data were the source for 2005 ozone-season heat rates and NOx emission rates. The increased cost of electricity generation under a higher emissions price relative to the base price for each generating unit i at a given hour was ΔCi: ΔCi = ci′ui′ − ciui

(2)

where for each generating unit i, ui and ui′ are its outputs in MWh under the base emissions price and higher emissions price, and ci and ci′ are its variable costs under the base emissions price and higher emissions price. The increased daily cost of time-differentiated policy was the summation of ΔCi over each hour of the day and all EGUs within the electricity grid. For the base case, NOx pricing was applied uniformly to all units in PJM with a price of $2k/ton, approximately reproducing the observed NOx price as of summer 2005.29 In operating a time-differentiated or any similar flexible NOx regulation that targets high ozone days, the system operator would set the following day’s NOx price based on the ozone forecast. In this study, we explored the impacts of increasing the NOx price to $30k/ton, $50k/ton, or $100k/ton on predicted high ozone days. It should be noted that the cost of abatement is not the NOx price times the total abatement. The NOx prices serve as signals of the relative scarcity. An equivalent system 5609

dx.doi.org/10.1021/es204290s | Environ. Sci. Technol. 2012, 46, 5607−5615

Environmental Science & Technology

Article

Figure 1. Decision framework for selecting time-differentiated NOx pricing in the presence of risks in ozone forecasts during an N-day period. H and L represent high and low ozone days, respectively. On each day, four possible outcomes are associated with the time-differentiated policy with probabilities denoted as p1, p2, p3, and p4.

A decision tree representation of this process is shown in Figure 1. The probability of a false positive (Type I error) or false negative (Type II error) are denoted as p and q:

simulated days using a two-state Markov Chain, where the states were H or L, representing high or low ozone days. A high (or low) ozone day was defined as a day in which the 8-h ozone concentrations at 30 out of the 37 ozone monitoring sites (80% of the sites) in the Philadelphia/Baltimore region was higher (or lower) than an 8-h averaged threshold concentration. Three possible thresholds were considered: 65, 75, or 85 ppb. During June 1 to Sept 30, 2002, there were 27 high and 95 low ozone days for the 75 ppb threshold, 54 high and 68 low ozone days for the 65 ppb threshold, and 8 high and 114 low ozone days for the 85 ppb threshold. The transition probability matrices for the two-state Markov Chain were estimated as

Pthresholdozone

(4)

p = Pr[forecast L|actual H]

(5)

The probabilities that a randomly chosen day was a high or low ozone day were defined as a and 1 − a, where a was a function of the transition matrix of the two-state Markov model and the initial state of the model. On each day, four possible outcomes were associated with the time-differentiated policy, with probabilities denoted as p1, p2, p3, and p4 and obtained as:

⎡ H L ⎤ H⎢ ⎥ = ⎢1 − π π ⎥ L⎢ ⎣ θ 1 − θ ⎥⎦

p1 = Pr[actual H] × Pr[forecast H|actual H] = a(1 − q) (6)

⎡ 0.72 0.28 ⎤ P65 = ⎢ ⎣ 0.22 0.78 ⎥⎦ ⎡ 0.52 0.48 ⎤ P75 = ⎢ ⎣ 0.15 0.85 ⎥⎦ ⎡ 0.25 0.75⎤ P85 = ⎢ ⎥ ⎣ 0.05 0.95⎦

p = Pr[forecast H|actual L]

p2 = (1 − a)q

(7)

p3 = aq

(8)

p4 = (1 − a)(1 − q)

(9)

Thus, 4N outcomes were associated with each time-differentiated scenario during an N-day ozone season. Mean outcomes were estimated by applying Monte Carlo simulation for a sample size, N, of 50 000 days. The days in the stochastic analysis were sampled from the set of modeled days June 1 to Sept 30, 2002, with replacement. The set of modeled days was sorted into low ozone and high ozone days, and as we simulated the Markov Chain, we chose first whether to sample a high or low ozone day and then drew one day randomly from the appropriate subset. On the ith day, the expected daily cost and the associated ozone reduction were:

(3)

where π denotes the conditional probability that a randomly chosen day was a low ozone day, given that the previous day was a high ozone day, and θ denotes the conditional probability that a randomly chosen day is a high ozone day, given that the previous day was a low ozone day. Modeled conditions were assumed to be the “actual” outcomes, and errors in ozone forecasts were simulated by sampling the forecast, conditional on the modeled ozone and a prescribed error rate. An analysis of the mean total cost and mean total reduction in peak ozone under each NOx pricing policy was conducted for different pairs of false positive and false negative error rates in ozone forecasts to determine threshold values of ozone forecast errors below which each time-differentiated policy was more cost-effective than a prescriptive strategy.

(Δcost)i = (p1 + p2 )Ui = [a(1 − q) + (1 − a)p]Ui (10)

(ΔO3)i = p2 Di = a(1 − q)Di

(11)

where Ui was the increased cost of electricity generation under a higher NOx price on the ith day of the N-stage decisionmaking process and Di was the associated reduction in the daily 5610

dx.doi.org/10.1021/es204290s | Environ. Sci. Technol. 2012, 46, 5607−5615

Environmental Science & Technology

Article

maximum 8-h ozone concentration. In this work, ΔO3 (ppb) was defined as: 37

ΔO3 = Mcj , k =

∑ (si ,j ,base − si ,j , k) i=1

(12)

where ⎧ ci , j , k − threshold, ci , j , k > threshold si , j , k = ⎨ ⎪ 0, otherwise ⎩ ⎪

for thresholds of 65, 75, or 85 ppb, Mcj,k is the monitor integrated ozone impact on day j under policy scenario k, ci,j,k is the daily maximum 8-h ozone concentration in grid cell i where the ith ozone monitoring station is located, and reductions are summed over the 37 ozone monitoring sites in the Philadelphia/Baltimore area. The mean Δcost and ΔO3 generated by the Monte Carlo simulation for the timedifferentiated scenarios were denoted as U1 and D1. The Δcost and ΔO3 under the technology-based scenarios were U2 and D2. The metric for comparing the time-differentiated and technology scenarios to the base case ($2k/ton) was the ratio of Δcost to ΔO3. Because U2 and D2 were not a function of p and q, the values for U1/D1 obtained by the stochastic model were used to compare to the deterministic U2/D2.



RESULTS Deterministic Scenarios: Seasonal NOx Pricing versus Technology-Based Policies. NOx price increases of $30k/ ton, $50k/ton, or $100k/ton were implemented throughout the ozone season to examine the magnitude of emissions and ozone reductions achievable with pricing versus technologybased strategies in the Classic PJM grid. Figure 2 shows total NOx emissions for an average electricity demand day and a peak demand day during the 2002 ozone season. As expected, NOx emissions decreased with increases in NOx price. More notably, even on a peak demand day and at the peak demand hours, sufficient flexibility existed in the electric grid to allow generation to be switched from high to low NOx emitting facilities if given an incentive through higher NOx prices. For most hours, SNCR controls resulted in reductions similar to the $50k/ton scenario. At peak demand hours (17:00), SNCR resulted in greater NOx reductions, but SNCR and the NOx pricing policies were generally similar in the midmorning when NOx emissions can influence peak ozone concentrations later in the day.38−41 Over the entire ozone season, the aggregate fraction of NOx reduced over the Classic PJM system relative to the base case was 23%, 30%, and 41% for the $30k/ton, $50k/ ton, and $100k/ton pricing policies, and 34% and 28% for the SNCR and SCR scenarios. Given that SCR installation at only 32% of coal-fired facilities was considered, it is expected that installation at all facilities would result in the greatest NOx emissions reductions. The emissions reduction under the pricing polices resulted only from substituting lower-emitting, higher-cost generation from what would have been used in a least-cost dispatch, primarily switching from coal generation to gas generation. Demand was held fixed, and no power was imported from other regions. With the exception of peak demand afternoons, dispatching electricity generation from high to low NOx emitting EGUs through NOx pricing is as effective as SNCR controls at all coal-fired EGUs or SCR controls at a subset of facilities, while meeting demand in the Classic PJM grid. PJM, as with other power systems, has

Figure 2. Time series of total NOx emissions on (a) June 2 (average demand day) and (b) Aug 12, 2002 (peak demand day) under NOx pricing of $2k/ton, $30k/ton, $50k/ton, and $100k/ton and with the installation of SNCR on all coal-fired units and SCR on 32% of the coal-fired facilities with the highest NOx emissions.

significant excess capacity for all except the peak demand hours of the year. In Section S5, Supporting Information, the average daily percentage of grid cells with Philadelphia/Baltimore ozone monitoring stations that have base case ozone concentrations of 75 ppb or greater during June to Sept 2002 and have increases or decreases in maximum 8-h averaged ozone concentrations are compared for seasonal NOx pricing and the implementation of SNCR or SCR. Although daily variation existed, SCR was associated with the largest spatial area (>90%) of reduced ozone concentrations in Philadelphia/Baltimore grid cells with monitors, but all scenarios offered substantial benefits. On average, seasonal NOx pricing of $30k/ton and SNCR reduced daily maximum 8-h ozone concentrations by at least 1 ppb over 14% of grid cells. Pricing at $50k/ton or $100k/ton or implementation of SCR at a subset of facilities reduced peak ozone concentrations by at least 1 ppb over 23%, 33%, and 55%, of Baltimore/Philadelphia grid cells with ozone monitors. Although all policy scenarios were associated with increased ozone concentrations (“hotspots”), these areas were compara5611

dx.doi.org/10.1021/es204290s | Environ. Sci. Technol. 2012, 46, 5607−5615

Environmental Science & Technology

Article

tively smaller spatially than those associated with reduced ozone concentrations. The average reduction in the predicted fourth-highest daily maximum 8-h averaged ozone concentration in the grid cells with the 37 Philadelphia/Baltimore monitoring stations, which is the metric used in determining attainment with the NAAQS, was greatest for SCR at 4.0 ppb followed by 1.2 ppb for the $100k/ton scenario, 0.8 and 0.7 ppb for the $50k/ton and SNCR scenarios, and 0.6 ppb for the $30k/ton scenario. None or only one grid cell (